An orbital data-center fleet supplying 10% of projected global artificial-intelligence data-center demand in 2030 would require approximately 700 or more large rocket launches every year under a scenario in a new Innovation for Cool Earth Forum working paper. Columbia University publicized the report on November 12, 2026.
The Space-Based Data Centers Roadmap, led by David Sandalow, examines the hardware and operating systems connecting computing capacity with satellite mass and launch activity. The calculation describes sustained deployment and replenishment using Starship-sized rockets. It is a modeled operating scenario.
The paper explains that radiator mass, chip operating temperature, and spacecraft mass per unit of computing capacity influence the fleet’s physical scale. Its authors include specialists in energy, engineering, security, and law. The November-dated document is identified as a working paper that has not undergone peer review.
Computing Hardware Becomes Spacecraft Hardware
An orbital data center combines processors and memory with the equipment needed to operate a satellite. Electricity supplies the computing equipment, communications move information between users and processors, and thermal hardware transfers waste heat away from electronics. These functions place computing demand inside a spacecraft’s mass and power budgets.
The relationship is physical. Electrical power used by computing equipment becomes heat that must leave the vehicle. Carrying more computing equipment can increase the electricity and cooling capacity required, adding equipment that must also be launched.
NASA’s spacecraft thermal-control reference explains why cooling in orbit differs from cooling a building. In a vacuum, there is no surrounding air to carry heat away by convection. Heat moves through structures and thermal connections inside a spacecraft, then leaves exposed surfaces through radiation.
A radiator is a surface designed to emit that energy. Its effectiveness depends on temperature, surface area, and the material’s ability to emit infrared radiation. Designers also account for incoming sunlight and heat arriving from Earth or nearby spacecraft components.
Surface treatments control these exchanges. A radiator coating can reflect much of the incoming sunlight and still emit thermal energy efficiently. Thermal straps provide conductive connections between hot equipment and other spacecraft structures. Deployable radiators provide additional emitting area after a spacecraft leaves its launch configuration.
These mechanisms explain why a cold space environment and a working cooling system are separate engineering concepts. The equipment transfers heat through a designed path from the processor to a surface that can radiate it.
The Terrestrial Demand Behind Orbital Proposals
The International Energy Agency’s 2026 energy and AI assessment projects global data-center electricity use increasing from 485 terawatt-hours in 2025 to approximately 950 terawatt-hours in 2030. In its central projection, AI-focused facilities increase their electricity consumption threefold during that period.
A terawatt-hour measures energy consumed over time. It differs from a gigawatt, which measures the rate of electricity use or supply. Computing performance is another quantity again: it describes the work performed by processors. Comparing proposed systems requires identifying which of these quantities a stated capacity represents.
The agency also reports that AI-server power density increased elevenfold between 2020 and 2025. Greater power density places more electricity demand and heat generation within each equipment rack. That development connects terrestrial computing trends with the power and thermal questions considered in orbital designs.
New Space Economy’s earlier coverage of orbital data-center programs describes the companies and development stages involved in this market. The ICEF paper adds a system-level treatment of how computing demand translates into spacecraft and recurring transportation requirements.
The report also examines debris, atmospheric effects, astronomy, and the legal treatment of large constellations. Its policy recommendations favor phased deployment and restrictions on expansion. Those recommendations are the authors’ proposals; the launch estimate is their quantitative scenario.
The paper is available through the ICEF roadmap repository. Its preface invites comments and corrections and states that sections may be updated as further research becomes available.
